tumours: initial results from the P-PRRT trial
Méthodes 79 patients avec TNE recevant une thérapie au 177 Lu-octreotate (279 cycles thérapeutiques) ont été inclus dans notre étude Les images QSPECT avec trois points dans
le temps (3TP : jour 0, 1 et 3) ont été effectuées après chaque administration thérapeutique. La dosimétrie a été obtenue à l’aide de petits volumes d’intérêt pour l’échantillonnage de la concentration d’activité pour le rein, la moelle osseuse et la tumeur avec la plus intense captation. L’exactitude de la dosimétrie simplifiée basée sur deux temps (2TP : jour 1 et 3,
régression monoexponentielle) ou sur une analyse ponctuelle (1TPD3 : jour 3) a été évaluée,
ainsi que celle des méthodes hybrides basées sur 2TP pour le premier cycle et 1TP (jour 1
ou 3; 2TP/1TPD1 et 2TP/ 1TPD3, respectivement) ou aucune imagerie (basée sur l’IA
seulement; 2TP/NI) pour les cycles d'induction subséquents. L'accord inter-observateur a
été évalué pour les méthodes 3TP, 2TP et hybrides 2TP/1TPD3 sur 60 cycles d'induction (15
patients). Le taux de filtration glomérulaire estimé (eTFG), les descripteurs corporels
(poids, surface corporelle (SC), poids corporel maigre (PCM)) et les produits des deux ont été évalués pour leur capacité de prédire la dose de radiation absorbée par le rein au premier cycle.
Résultats Les estimations de la dosimétrie sur 2TP sont fortement corrélées avec celles des
données obtenues par 3TP pour tous les tissus (Spearman r > 0,99, P < 0,0001). Les erreurs relatives entre les méthodes sont petites, en particulier pour le rein et la tumeur, pour lesquels les erreurs relatives médianes ne dépassent pas 2 %; les intervalles interdéciles s'étendent sur moins de 6 % et 4 %, respectivement, pour les estimations par cycle et
108
cumulatives. Pour la moelle osseuse, les erreurs ont été légèrement plus élevées (erreurs médianes < 6 %, intervalle des interdéciles < 14 %). Globalement, la force des corrélations des doses absorbées estimées par des méthodes simplifiées comparées à celles obtenues par la méthode 3TP a tendance à diminuer progressivement, et les erreurs relatives à
augmenter, dans l'ordre suivant: 2TP, 2TP/1TPD3, 1TPD3, 2TP/1TPD1 et 2TP/NI. Pour la
tumeur, le scénario de 2TP/NI était très inexact en raison de l'interférence de la réponse thérapeutique. Il y avait un excellent accord inter-observateur entre les trois observateurs, en particulier pour la dose rénale estimée à l'aide des méthodes 3TP et 2TP, avec des erreurs moyennes inférieures à 1 % et des écarts-types de 5 % ou moins. Les produits TFG•PCM et TFG•SC ont prédit le mieux rapport entre l'AI et la dose rénale (GBq/Gy) pour le premier cycle (Spearman r = 0,41 et 0,39, respectivement; P < 0,001). Pour le premier cycle, l'AI personnalisée proportionnelle à TFG•PCM ou TFG•SC a diminué l’intervalle de dose de radiation absorbée rénale délivrée entre les patients par rapport à l'AI empirique. Pour les cycles subséquents, l'AI personnalisée optimale pourrait être déterminée de façon fiable en fonction du GBq/Gy rénal du cycle antérieur avec une erreur de moins de 21 % chez 90 % des patients.
Conclusion Un protocole de dosimétrie simplifiée basée sur les images QSPECT à deux
temps aux jours 1 et 3 après la PRRT fournit des estimations de dose reproductibles et plus précises que les techniques basées sur un seul point temporel pour les cycles subséquents. En effet, cette méthode cause des inconvénients limités au patient par rapport aux protocoles impliquant l’ajout des points temporels plus tardifs. La dose de radiation rénale absorbée rénale cumulative peut être normalisée en personnalisant l'AI en fonction du produit du TFG avec PCM ou SC pour le premier cycle et de la dosimétrie rénale antérieure pour les cycles subséquents.
109
Abstract
Background Routine dosimetry is essential for personalized 177Lu-octreotate peptide
receptor radionuclide therapy (PRRT) of neuroendocrine tumours (NETs), but practical and robust dosimetry methods are needed for wide clinical adoption. The aim of this study was to assess the accuracy and inter-observer reproducibility of simplified dosimetry protocols based on quantitative single-photon emission computed tomography (QSPECT) with a limited number of scanning time points. We also updated our personalized injected activity (IA) prescription scheme.
Methods Seventy-nine NET patients receiving 177Lu-octreotate therapy (with a total of
279 therapy cycles) were included in our study. Three-time-point (3TP: Day 0, 1 and 3) QSPECT scanning was performed following each therapy administration. Dosimetry was obtained using small volumes of interest activity concentration sampling for the kidney, the bone marrow and the tumour having the most intense uptake. Accuracy of the simplified dosimetry based on two-time-point (2TP: Day 1 and 3, monoexponential fit) or a single- time-point (1TPD3: Day 3) scanning was assessed, as well as that of hybrid methods based on 2TP for the first cycle and 1TP (Day 1 or 3; 2TP/1TPD1 and 2TP/1TPD3, respectively) or no imaging at all (based on IA only; 2TP/NI) for the subsequent induction cycles. The inter-observer agreement was evaluated for the 3TP, 2TP and hybrid 2TP/1TPD3 methods using a subset of 60 induction cycles (15 patients). The estimated glomerular filtration rate (eGFR), body size descriptors (weight, body surface area (BSA), lean body weight (LBW)), and products of both were assessed for their ability to predict IA per renal absorbed dose at the first cycle.
Results The 2TP dosimetry estimates correlated highly with those from the 3TP data for all
tissues (Spearman r > 0.99, P<0.0001) with small relative errors between the methods, particularly for the kidney and the tumour, with median relative errors not exceeding 2% and interdecile ranges spanning over less than 6% and 4%, respectively, for the per-cycle and cumulative estimates. For the bone marrow, the errors were slightly greater (median errors <6%, interdecile ranges <14%). Overall, the strength of correlations of the absorbed
110
dose estimates from the simplified methods with those from the 3TP scans tended to progressively decrease, and the relative errors to increase, in the following order: 2TP, 2TP/1TPD3, 1TPD3, 2TP/1TPD1 and 2TP/NI. For the tumour, the 2TP/NI scenario was highly inaccurate due to the interference of the therapeutic response. There was an excellent inter-observer agreement between the three observers, in particular for the renal absorbed dose estimated using the 3TP and 2TP methods, with mean errors lesser than 1% and standard deviations of 5% or lower. The eGFR•LBW and eGFR•BSA products best predicted the ratio of IA to the renal dose (GBq/Gy) for the first cycle (Spearman r = 0.41 and 0.39, respectively; P < 0.001). For the first cycle, the personalized IA proportional to eGFR•LBW or eGFR•BSA decreased the range of delivered renal absorbed dose between patients as compared with the fixed IA. For the subsequent cycles, the optimal personalized IA could be determined based on the prior cycle renal GBq/Gy with an error of less than 21% in 90% of patients.
Conclusion Simplified dosimetry protocol based on two-time-point QSPECT scanning on Day 1 and 3 post-PRRT provides reproducible and more accurate dose estimates than the techniques relying on a single time point for non-initial or all cycles, and results in limited patient inconvenience as compared to protocols involving scanning at later time points. Renal absorbed dose over the four-cycle induction PRRT course can be standardized by personalizing IA based on the product of eGFR with LBW or BSA for the first cycle, and on prior renal dosimetry for the subsequent cycles.
Keywords Dosimetry, neuroendocrine tumours, peptide receptor radionuclide therapy,
111
Background
For patients with metastatic neuroendocrine tumours (NETs), peptide receptor radionuclide therapy (PRRT) with 177Lu-octreotate is an effective palliative treatment that rarely causes serious toxicity [1, 2]. PRRT has been mostly administered as a 4-cycle induction course using a fixed injected activity (IA) of not more than 7.4 GBq per cycle, in order to not exceed cumulative absorbed doses of 23 Gy to the kidney and 2 Gy to the bone marrow (BM) in the majority of patients [1-4]. However, it is well known that for these critical organs, and in particular for the kidney which is the dose-limiting organ for most patients, the absorbed dose per IA is highly variable and usually lower than 23 Gy per 4 cycles, resulting in most patients being undertreated with such an empiric PRRT regime [5, 6]. We and others have proposed personalized PRRT (P-PRRT) protocols in which the number of fixed-IA cycles or the IA per cycle are modulated to deliver a safe prescribed renal absorbed dose, with the aim to maximize tumour irradiation while keeping the toxicity low [4, 6]. Such P-PRRT protocols require careful dosimetry monitoring, which is often perceived as a complex and resource-consuming process, therefore constituting a barrier for wide clinical adoption. As a result, the clinical practice of “one-size-fits-all” PRRT prevails, at the potential cost of delivering a suboptimal treatment to most patients.
We have routinely been performing post-PRRT dosimetry using quantitative single- photon emission computed tomography (QSPECT) combined with the small-sphere volume of interest (VOI) activity concentration sampling [5, 6]. Aiming to simplify the dosimetry process and to reduce the clinical burden thereof, we examined the impact of reducing the number of QSPECT sessions on the accuracy and the inter-observer reproducibility of the resulting dose estimates. In parallel, based on a large dataset from our growing cohort of patients treated with PRRT, we updated our personalized IA determination scheme.
112
Methods
Patients and PRRT cycles
From November 2012 to December 2017, 81 patients with progressive metastatic and/or symptomatic NET were treated with PRRT at CHU de Québec – Université Laval. Two patients who underwent only one cycle each were excluded because of their incomplete dosimetric data, and therefore only data from 79 patients was analysed. This includes 23 patients who received only empiric PRRT (i.e. fixed IA of approximately 8 GBq, occasionally reduced) until March 2016, for whom the requirement for consent was waived due to the retrospective nature of the analysis. All other patients were enrolled in our P-PRRT trial (NCT02754297) and gave informed consent to participate (protocol described in [6]). Patient characteristics are reported in Table 1.
113
Table 3 - 1 Patient characteristics
All patients (n=79) Gender, n (%) Female Male 36 (45.6) 43 (54.4)
Age at first cycle, median (range) 60.7 (26.1-82.3)
Site of primary tumour, n (%)
Small intestine Pancreas Adrenal gland a Lung Colon Stomach Esthesioneuroblastoma Unknown 30 (38.0) 26 (32.9) 6 (7.6) 6 (7.6) 2 (2.5) 1 (1.3) 1 (1.3) 7 (8.9) Metastases, n (%) Liver Lymph nodes Bone Lung Other b 66 (83.5) 51 (64.6) 29 (36.7) 9 (11.4) 25 (31.6)
Body size descriptors, mean SD (range)
Weight (Kg)
Lean body weight (Kg) Body surface area (m2)
72.1 16.6 (42.6 – 121.0) 52.4 9.9 (35.4 – 81.2)
1.8 80.2 (1.4 – 2.5)
eGFR (ml/min/1.73 m2), mean SD (range) 86.3 22.2 (42.0 – 154.1)
Number of cycles, n (%) 1 2 3 4 5 6 7 8 8 (10.1) 6 (7.6) 16 (20.3) 38 (48.1) 3 (3.8) 6 (7.6) 1 (1.3) 1 (1.3) Type of cycles, n (%) Empiric only Personalized only Mixed 23 (29.1) 45 (57.0) 11 (13.9) eGFR, Estimated glomerular filtration rate; PRRT, Peptide receptor radionuclide therapy
a Three patients with pheochromocytoma and three patients with paraganglioma b Peritoneum, ovary, subcutaneous, pleura, meninges
114
Two hundred and eighty-four therapy cycles were administered during the study period. Five cycles in five patients were excluded from the analysis because of dosimetry protocol deviation or missing data. Among the 279 therapy cycles analysed, 142 were empiric (median IA = 7.6 GBq; range, 3.8-9.1 GBq) and 137 were personalized (median IA = 9.0 GBq; range, 0.7-32.4 GBq). Anti-nausea premedication (ondansetron and dexamethasone) and a nephroprotective amino acid solution (lysine and arginine) were administered [1]. We administer a four-cycle induction course for which the prescribed cumulative renal dose is 23 Gy (5 Gy at the first cycle; two-monthly intervals) and, in responders only, we offer consolidation, maintenance and/or salvage cycles (prescribed renal dose of 6 Gy each; personalized intervals). As previously described, prescribed renal absorbed radiation doses were reduced in patients with renal or bone marrow impairment [6].
Reference dosimetry method
At each cycle, after therapeutic administration of 177Lu-octreotate,
QSPECT/computed tomography (QSPECT/CT) scans were performed at approximately 4 hours (Day 0), 24 hours (Day 1) and 72 hours (Day 3) using a Symbia T6 camera (Siemens Healthcare, Erlangen, Germany) (Fig. 1) [6, 7]. Following the same data processing as described in [7], the dead-time corrected reconstructed images were converted into the positron emission tomography (PET) DICOM format, which includes a “rescale slope” parameter that converts count data into Bq/mL and also enables display of QSPECT images in standardized uptake values normalized for body weight (SUV).
115
Fig. 3 - 1
Post-treatment serial QSPECT/CT was performed at (from left to right) 5, 24 and 70 h after a 22.0 GBq 177Lu-octreotate administration in a 55-year-old male with metastatic NET of unknown origin. Small volumes of interest (2-cm diameter) were placed over tissues of interest. Left kidney (red arrows), L5 bone marrow cavity (orange arrows) and dominant tumour (green arrows) VOIs are pointed on anterior maximum intensity projections (top row) and selected transaxial fusion slices (mid and bottom rows). QSPECT images are normalized using an upper SUV threshold of 7. During this consolidation cycle, the personalized injected activity allowed the delivery of 6.1 Gy (6.0 Gy prescribed) to the kidney.
116
These three imaging time points were initially selected for the following practical reasons: (1) the Day 0 scan does not incur any additional hospital visit for the patient and allows capturing early kinetics of the radiopharmaceutical; and (2) performing scans beyond Day 3 would not be easy for logistical reasons (PRRT being administered on Tuesday, Day 4 or 5 would fall on the weekend) and would inconvenience patients (in particular those out-of-city patients who would need to prolong their stay).
As in our clinical practice we routinely performed dosimetry based on the data acquired at these three time points (3TP), this approach constituted the reference method for the present analysis. In brief, at each time point, we sampled the activity concentration in tissues of interest (Fig. 1), including both kidneys (areas of representative parenchymal uptake), the BM (L4 and L5 vertebral bodies, or elsewhere when the latter were obviously
affected by metastases) and the tumour having the most intense uptake (Tumourmax), using
2-cm (4.2 cm3) spherical VOIs, as previously described [5, 6]. This was performed using
either Hybrid Viewer (Hermes Medical Solutions, Stockholm, Sweden) or MIM Encore (MIM Software Inc., Cleveland, OH, USA) software. As previously described in [6], we also computed the total body retention for the purpose of computing the cross-dose
component of the BM absorbed dose (BMcross), which we added to the self-dose component
(BMself) to estimate the total BM absorbed dose (BMtotal).
Based on these 3TP data, trapezoidal-monoexponential (3TPTM) time-activity curves
(TACs) were drawn using the following procedure (Fig. 2). For each organ/tumour, a constant mean SUV was assumed from the time of 177Lu-octreotate injection until the time of the Day 0 scan (approximately 4 hours). This was followed by a linear (trapezoid) fit to the SUV corresponding to the Day 0 and the Day 1 scans. Then, a monoexponential curve was fit using the Day 1 and Day 3 data, resulting in an effective decay model being used
from Day 1 onwards (trapezoidal-monoexponential; 3TPTM; Fig. 2). However, in cases
when the Day 3 SUV was higher than that corresponding to Day 1, we assumed a linear
SUV variation between Day 1 and 3, followed by the physical decay of activity (i.e. biol =
117 Fig. 3 - 2 A. B. C. D
Time-activity curves (TACs) of the renal (A), tumour (B) and bone marrow (C) activity concentrations, and of the whole- body retention (D) over time for the patient case illustrated in Fig. 1. TACs in MBq/cc or MBq (red) and SUV or percentage of injected activity (%IA) (blue) are illustrated for the three-time-point (3TP, solid lines) and two-time-point (2TP, dashed lines) method.
Then, the area under each TACs curve was integrated and multiplied by the appropriate activity concentration dose factors (ACDF). The values of these factors have been derived from OLINDA/EXM software data (Vanderbilt University, Nashville, TN), as previously described [6]: 84 mGy⋅g/MBq/h for Tumourmax and 87 mGy⋅g/MBq/h for kidneys and BMself. For BMcross, we integrated the total body activity over time and multiplied it by a dose factor of 1.09 × 10-4 mGy/MBq/h for males or 1.29 × 10-4
118
mGy/MBq/h for females, i.e. to account for their different gamma fraction of energy deposition from the whole body to the BM [6].
Simplified dosimetry methods
From our experience and as suggested by others, the Day 0 data, although it captures the rapid kinetics of the radiopharmaceutical (which includes competing accumulation and rapid washout), contributes little to the area under the TAC, which is mostly determined by the slow washout kinetics and tends to follow a monoexponential decay beyond 24 hours [8]. Accordingly, we eliminated the Day 0 data from all simplified dosimetry approaches. A total of five methods were investigated, as detailed below.
2TP Two-time point (2TP) dosimetry estimates were obtained using VOI data from
Day 1 and Day 3 scans. From time of administration to infinity,
monoexponential (2TPM) effective decay was applied, except in cases of
biological accumulation of activity, i.e. when the SUV of the tissue increased between Day 1 and Day 3. In such cases, we assumed a SUV equal to that of Day 3 SUV (biol = 0), from time of treatment administration to infinity, and thus applied only physical decay (eff = phys; 2TPC). The 2TP method is the
combination of 2TPM and 2TPC.
1TPD3 As proposed by Hänscheid et al., we estimated doses using a single-time point
method based on the Day 3 data (1TPD3) [8]. In this method, the activity
concentration (MBq/cc) was multiplied by the time at which the Day3 scan was performed (h) and by 0.25 Gy⋅g/MBq/h (based on Eq. 8 in [8]). To compute BMcross, the total whole-body activity (MBq) was multiplied by imaging time (h) and by 3.2 × 10-7 Gy/MBq/h for males or 3.7 × 10-7 Gy/MBq/h for females, i.e. the gamma fraction of energy deposition from the whole body to the BM, multiplied by 0.25 Gy⋅g/MBq/h (from [8], as above), divided by 87 mGy⋅g/MBq/h (ACDF of the BM and kidney).
119
2TP/1TPD1 We evaluated a hybrid dosimetry protocol based on the 2TP method for the first cycle, as described above, and employed a single scan on Day 1 for the subsequent induction cycles. In this scenario, the absorbed dose to a given tissue during the second and the subsequent induction cycles was obtained by applying the monoexponential curve corresponding to the effective decay, as determined for this tissue during the first cycle, to the activity concentration observed on Day 1 of the subsequent cycle.
2TP/1TPD3Same as 2TP/1TPD1, but the single scan on subsequent cycles was that performed on Day 3.
2TP/NI Similar to the two previous methods, this method was also based on 2TP scanning for the first cycle, but no imaging (NI) was performed for the subsequent cycles. For the latter, the absorbed dose per IA during the subsequent cycles was simply assumed to be equal to that delivered during the first cycle.
Cumulative renal, BMtotal and Tumourmax absorbed doses were compiled for all patients who received three or four induction cycles (n=65). Per-cycle and cumulative doses resulting from each of the simplified dosimetry methods were compared with those obtained using the reference (3TP) method, and relative errors were calculated.
Inter-observer variability
For 60 induction cycles in 15 patients, the dosimetry analysis was performed independently by three observers having different backgrounds and purposely varied levels of experience in internal dosimetry. Observer 1 (M.D.P.), a certified endocrinologist, current PRRT Fellow and Ph.D. student, performed 258 of the 279 primary analyses described in this paper and, as such, accumulated the most experience with this dosimetry procedure. Observer 2 (F.A.) was a certified Nuclear Medicine Physician and current Nuclear Oncology Fellow who performed 21 primary analyses. Observer 3 (N.S.) was an M.D. student who was new to both nuclear medicine and dosimetry and who received only
120
a short training. Relative errors of per-cycle and cumulative absorbed doses between each pair of observers were computed for the reference method (3TP), and the two most accurate simplified methods.
Personalized 177Lu-octreotate activity prescription
We previously derived a model based on the body surface area (BSA) and the estimated glomerular filtration rate (eGFR, according to the CKD-EPI Creatinine Equation [9]) to determine the personalized 177Lu-octreotate activity to be administered at the first cycle [6]. Using data from our entire cohort of 79 patients, we aimed to formulate a simpler prescription equation. To this end, we correlated the ratio of IA to the renal absorbed dose estimated from the first cycle (GBq/Gy, obtained by the 3TP or the 2TP methods) with the patient’s weight, lean body weight (LBW), BSA, eGFR and the products of eGFR with each of the three body size descriptors. Then, for each of these seven correlations, we performed a linear regression forced through the origin (eliminating the intercept) and calculated the relative errors of the predicted renal GBq/Gy using the slope of the linear regression.
We also compared the accuracy of predicting the renal GBq/Gy in any given non- initial cycle with that from the previous cycle or with the average renal GBq/Gy of the two previous cycles, as we have initially been doing in our P-PRRT trial [6].
Statistical methods
Data are presented as median and interdecile range or as mean ± SD according to the data distribution using D’Agostino-Pearson omnibus normality test. Ranges are also reported. Pearson or Spearman correlations were used depending on the normality of the data. A difference was considered as statistically significant if the P-value was below 0.05. Correlations and linear regressions were performed using GraphPad Prism software